Analyzing Audio Patterns During the 2024 Total Solar Eclipse

Kaitlin Heintzman

St. Lawrence University

2025-04-25

Outline

  • Introduction

  • Data Collection

  • Data Frame Creation

  • Modeling

  • Conclusion

Introduction

  • Our goal was to look deeper into effects that the total solar eclipse has on wildlife activity.

  • Analyzed audio data using 5 common audiological indices

  • Used Generalized Additive Models to produce visuals to search for temporal patterns

  • This was an amazing and unique opportunity to look into such a rare phenomenon and to use data that was collected by SLU faculty.


  • Inspiration for this analysis was taken from:

Gerber, J. E., Howard, D., & Quinn, J. E. (2020). Soundscape shifts during the 2017 total solar eclipse: An application of dispersed automated recording units to study ephemeral acoustic events. Biodiversity, 21(1), 41–47. https://doi.org/10.1080/14888386.2020.1715834

2024 Total Solar Eclipse

  • April 8th, 2024

  • Partial Eclipse from 2:11 to 4:35

  • Totality from 3:23 to 3:27

Image

Audio Recorder Placement

Data Collection

  • Recorders collected audio from March 30th to April 16th
Times Reasoning
5:45 - 7:15 30-45 minutes before sunrise
14:00 - 16:00 full period of time corresponding to the eclipse
19:00 - 20:00 ~30 minutes before and after sunset
23:00 - 23:30 sample nocturnal activity


  • Recorders sampled 55 seconds on and 5 seconds off

  • 20 audio recorders produced a total of 115,610 audio clips (>1 TB)

Audio Indices

  • Bioacoustic Index - Assesses diversity and abundance of biological noise. Gives higher value if there are more different types of sounds in different frequencies (2-8kHz)

  • Biophony - Measures the amount of sound within a certain frequency band

  • Acoustic Complexity - Focused on expressing the spatial and temporal complexity in sound by capturing the intensity and frequency shifts across an audio file.

  • Acoustic Diversity - Generates proportions over set intervals to assess evenness and variety in sound distribution

  • Acoustic Evenness - Measure how sound is distributed across frequency bands, assess equality and inequality of the rate that sound is produced from a source


Villanueva-Rivera LJ, Pijanowski BC (2018). soundecology: Soundscape Ecology. R package version 1.3.3, https://CRAN.R-project.org/package=soundecology.

Visualize and Youtube

https://youtu.be/3dftvaUn--Q

Using R and HPC

  • Function to take sound files and return cleaned data frame with information needed for analysis
AudioRecorder time day hour bioacoustic biophony
"/A001_SD001" 2024-03-31 15:07:00 2024-03-31 15:07:00 0.8716146 1.326163
"/A001_SD001" 2024-03-31 15:08:00 2024-03-31 15:08:00 0.6645511 1.887507
"/A001_SD001" 2024-03-31 15:09:00 2024-03-31 15:09:00 0.5972896 1.345134
"/A001_SD001" 2024-03-31 15:10:00 2024-03-31 15:10:00 0.5475493 1.548047
"/A001_SD001" 2024-03-31 15:11:00 2024-03-31 15:11:00 0.5584277 1.338760
"/A001_SD001" 2024-03-31 15:12:00 2024-03-31 15:12:00 0.3720922 1.718347


  • Over a terabyte of data

  • Due to massive size, we used St. Lawrence’s HPC (High-Performance Computer) to run and store our files incrementally.

  • With the HPC, each folder of clips took 10.5 hours to process.

Generalized Additive Modeling

Why GAM?

We can maintain additivity while incorporating non-linear functions!


\[ Index \sim spline(hour, by = day) + audio\_recorder+ day\]

  • Fit a smoothing function for time, and have it vary by day

  • Account for Audio Recorder differences, but treat this as a random effect

  • Adjust for overall differences in day


  • Numerically, the estimated degrees of freedom provided a comparison of line curvature between the days

Bioacoustic Model Visualization

Biophony Model Visualization

Findings

  • Based on our analysis we observed potential patterns in 3 of our indices:

    • Bioacoustic Index

    • Acoustic Evenness

    • Acoustic Diversity


  • These indices showed increases or decreases around the time of the eclipse, specifically with variation around the time of totality.

  • This suggests that wildlife is affected by the sudden changes in light that occurred during the solar eclipse or by some other eclipse-driven change.

Future Projects

  • Accounting for temporal autocorrelation: (Emilia Agostinelli ’26)

    • Adding more to the final model

    • Autocorrelation between audio clips


  • What species are being affected: (Aidan Fauth ’26)

    • Identifying the wildlife present in the clips

    • Machine learning possibility

Acknowledgements

  • Thank you to Erika Barthelmess, Jessica Harman (’27), Evelyn Albrecht (’25), and Kelsey Simler (’25) who made this project possible by collecting the data.

  • Thank you to the MCSS department and specifically Matt Higham for guiding me through this project.

  • Special Thanks to Ed Harcourt and Lisa Torrey for assistance in using and understanding the HPC process.